The Research Of Ultrafast Ultrasound Imaging Method And Its Implementation On CUDA

Ultrafast ultrasound imaging is a high frame rate ultrasound imaging modality, which is characterized by plane wave emission, coherent compounding and software parallel processing with wide application prospect. It has been applied in Doppler flow imaging and shear wave elasticity imaging, however, the application and commercialization are restricted by signal bandwidth and backend computing capability problems. Consequently, it has important research significance for the research of imaging method and the realization of high speed parallel processing.Above all, the structure of ultrafast ultrasound B mode imaging system is introduced and according to its characteristics, the field II ultrasound simulation platform of plane wave sound field simulation, concept of coherent compounding, and basis of GPU general purpose parallel calculations are followed.Secondly, in terms of the theory of imaging methods, they are intr oduced from two aspects, the time-domain coherent plane wave compounding method and Frequency-domain Fourier and F-K migration method. Time domain method is based on geometric relationships in ultrasound scattering while frequency domain method is mainly based on wave equation. The difference between two frequency-domain methods lies in the fact that the Fourier based method established the ultrasound scattering model according to the linear theory of ultrasonic system, but F-K migration simplifies ultrasonic pulse-echo process, and plane wave Exploding Reflector Model is established, based on which solving the ultrasound imaging problem.According to their own basic theory, this paper designs simulation process for each imaging method, and experiments via Matlab are implemented with RF data from multi-angle plane wave emission obtained by Field II. After that, how coherent compounding and number of compounding angles do impact on resolution and contrast of image is explored, by considering the comparison bet ween image quality and running time of three methods, F-K migration is the best of them.In the end, this paper designs and optimizes parallel programs of three imaging methods, integrating all the functions to minimize the switching time between kernel functions and pre-processing the coefficient offline to reduce real-time computation of parallel programs. After the implementation of parallel algorithms in kernel functions and verification of their correctness, total running time of each method is tested and F-K migration method appears to be the fastest. With the current hardware platform and detection area, actual imaging system can reach hundreds of frame rate.